Satellite-based estimation of ambient ammonia reveals meteorological influences and emission inventory uncertainties in South Korea.

Abstract

Atmospheric ammonia (NH3) is an important precursor gas of secondary PM2.5; however, sparse ground-based NH3 monitoring limits the characterization of its spatiotemporal distribution and provides insufficient observational evidence for evaluating emission inventories. To address these gaps, this study developed a spatiotemporal model to estimate monthly ground-level NH3 concentrations at 15 km resolution across South Korea for 2013-2017. A two-stage framework combining a linear mixed-effects model (LMM) and a generalized additive model (GAM) was applied to refine NH3 information from Cross-track Infrared Sounder satellite observations. The LMM incorporated meteorological variables and the Clean Air Policy Support System emission inventory, while the GAM characterized residual spatial patterns not fully represented by these predictors. The model demonstrated robust performance, with cross-validation R2 = 0.73, mean absolute error = 0.15, and root mean squared error = 0.21. Estimated NH3 concentrations were highest in agricultural areas, increasing from March to June, then declined. In the LMM, all meteorological factors were significantly associated with monthly NH3 concentrations. Temperature showed the strongest association with NH3 from March to June, peaking in June (+18.9% per +1 °C), while relative humidity and wind speed had their largest effects in March (+2.1% per +1% RH and -18.7% per +1 m/s). The GAM captured month-specific LMM residual patterns and identified agricultural NH3 hotspots that may reflect emission inventory gaps. These findings improve understanding of meteorological and emission-related controls on NH3 concentrations and support refinement of emission inventories, agricultural hotspots identification, and improved future PM2.5 air pollution assessment under changing environmental conditions.